Analysis of the perspectives of multiscale simulation of ... · Materials and device modeling It is...

Università di PisaMULT.EU.SIM. Workshop, Tenerife, 24/11/2011

M. MacucciDipartimento di Ingegneria dell’Informazione, Università di Pisa,

Via G. Caruso 16, I-56122, Pisa, Italy

[email protected]

Analysis of the perspectives of multiscale simulation of devices and

circuits


Summary

Structure and scope of the report that is being prepared for Mult.Eu.Sim on multi-scale device and circuit modelingProcedure followed to create a document representative of a shared viewSummary of the main contents and the main issues that have been raisedWhat may be interesting for the industryConclusions


DrainSource

Scope of the document

L

Shallow trench isolation

The scope of the document has been the role of modeling and, in particular, of multi-scale modeling of devices and circuits in the development of nanoelectronics in the near future

We have focused on the contribution that modeling can provide to developers of new technologies, device and circuit designers, and other researchers in the field

We have tried to understand what the industry currently lacks and what it will find valuable in the next few years


Structure of the document

The structure of the document is relatively straightforward:

Discussion of the relationship between device and material modeling and of the multi-scale structure that is involved

A few case studies

Relevance of high-performance computing and computational architectures

Role of physics based modeling

Simulation of time-dependent behavior

Relationship between device and circuit modeling


Origin of the document


Feedback from Europe

The response has been overwhelming, with comments and new contributions that have significantly enlarged and improved the document


Feedback from outside Europe


Materials and device modeling

It is recognized that, as device sizes are scaled down, the gap between the simulation of materials and that of devices is closing and in some cases they already overlapMulti-scale approaches are needed because ab-initiomethods cannot directly allow treatment of complete devices and circuitsHowever significant progress has been made with the development of linear-scaling codes, such as ONETEP and SIESTAThe requirement of self-consistency makes the computational burden very heavy


Organic Semiconductors

Disorder plays an important role in transport in organic semiconductors, and this adds to the computational complexity of its simulation

Recent multi-scale methods have been developed, which include treatment of polarontransport using first principles and a Kubo approach for quantum transport

Open issues are the interplay of impurity scatterings with phonons and the transport regime as a function of temperature


Nanowire transistors

Another field in which multi-scale modeling is expected to represent offer substantial contributions is that of nanowire transitors and of the circuits based on them

It has been demonstrated that a fully-functional “gate-all-around” transistor based on a silicon nanowire can operate correctly down to a channel length of 3 nm

An explicit quantum-mechanical description of the atomic scale structure has enabled calculations showing that operation is possible down to a diameter of 1 nm


Interactions between nanostructures and biological molecules

The comparable length scale of biomolecules and currently manufacturable semiconductor nanostructures open a completely new horizon of possibilities

This is true both in terms of exploiting the properties of biomolecules to help develop bottom-up fabrication approaches and of taking advantage of their recognition and biocatalytic properties to create “lab-on-a-chip” systems

Key to these studies is developing a multi-scale simulation framework that can handle the large simulations needed to atomistically describe the combined nanowire/biomoleculesystem, yet be able to treat the electronic structure of these systems to describe interactions with light and electronic transport phenomena


Current collapse in GaN HEMTs

GaN HEMTs are very promising devices for high-power, high-frequency applications, due the ability to operate safely at temperatures much higher than other semiconductor devices

However, there is still a reliability problem, because they often fail in a relatively short time if operated at high power levels

In particular, efforts are ongoing to understand the reasons of such a failure and of the “current collapse” effect observed after the application of an elevated drain-to-source voltage


Need for a multi-scale treatment of current collapse

Monte Carlo Device level Modeling

ShortTime Scale~ ps’s

Calibrate SentaurusHydrodynamic model

Hydrodynamic modeling with Sentaurus

LongTime Scalems‐s

Self‐Consistent Monte Carlo

device simulator for the Electron and Phonon

bath

Device level

Finite Volume Strain

Calculation

Atomistic band structurecalculation

atomistic level

Determine changes in

bandstructuredue to pits and cracks

DIFFERENT TIME SCALES

DIFFERENT LENGTH SCALES

Use deformation potential theory to calculate scattering

Supply charge density and Joule heating terms

Supply local strains and stresses

1.Trapping/detrapping2.Trap generation by hot carriers

Device and circuit level reliability modeling

LongestTime Scaleyears


Role of high-performance computing

The use of High Performance Computing is essential for achieving advances in frontier developments in the fields of first-principles calculations and multi-scale methodologies.

There is therefore a need for the usage of supercomputers, but the power of current supercomputers is not expected to become available to the majority of users in the foreseeable future

If we want the new hierarchies of multi-scale tools to be useful for device and circuit designers, they must run efficiently on widely available machines


Computational equipment (I)The best performance to cost ratio is currently achieved with GPUs (Graphic Processor Units) that were originally developed to deliver optimized 3D graphics but that, due to the massively parallel structure, can be exploited also for the solution of challenging numerical problems

The acceptance of GPUs for HPC is reflected by the fact that leading edge machines in the top 500 supercomputers contain GPUs

GPUs can handle hundreds of concurrent lightweight threads compared to the much smaller number of more complex threads that can be treated with CPUs


Computational equipment (II)High degrees of parallelism (although at least oenor two orders of magnitude below those of GPUs) can be achieved also with more conventional multi-core CPUs

The choice between GPUs and CPUs depends on the specificity of the algorithm and a well-balanced, general-purpose machine should contain the right mix of GPUs and CPUs

Extreme parallelization requires a lot of work on the software development front, from which very significant speedups can result


Predictive capabilities of physics-based modeling

There are success stories about the predictive capabilities of advanced numerical simulationsOne example is the prediction, resulting from an accurate interface model, that the TMR (Tunneling Magneto Resistance) ratio was going to improve dramatically when the dielectric was crystalline instead of amorphous Today TMR-based heads are equipping virtually all hard disks and have allowed a significant improvement in data storage density


Time-domain simulation

Time-domain behavior of nanodevices has received so far much less attention compared to DC transportThere is now a growing interest in the behavior as a function of frequency and in the noise characteristics of nanodevices, which requires an understanding of time-dependent transportThere are a few proposals in the literature that could be developed into full-fledged multi-scale tools, but some effort is needed, also in collaboration with experimental groups


Device and circuit levels

At the nanoscale the distinction between device modelling and circuit modelling starts to vanishOne of the main reasons is apparent in single-electronics: a SPICE description fails unless the capacitances at the nodes corresponding to the devices terminals are not significantly larger than the internal device capacitancesIf a model based on the I-V characteristics at the terminals fails, traditional circuit simulation cannot be performed and different approaches, based on multi-scale techniques have to be developed


What is interesting for the industry?(statements collected from industry representatives)

The industry is interested in simulation tools that can tell when something is not going according to intuition rather than in tools that are capable of very accurate predictions

The industry needs new tools very soon, because there are a lot of effects becoming important also in current generations of CMOS devices that cannot be properly dealt with existing software

The industry needs simulation tools that are very simple to use and in which all challenging setup tasks are performed automatically


Conclusions

Multi-scale modeling is the way to go in the next future for material-device-circuit simulation Big challenges are represented by the computational burden, the accessibilty to designers, the validation with experimental testsFuture multi-scale tools will need to exploit all the computational power of low-cost solutions (such as desktops integrated with GPU boards)A shared view of the device and circuit modeling community will be available in the report that is being prepared


Many thanks to:Frank Ortmann

Stephan Roche

Jim Greer

Georg Huhs

Xavier Oriols

Thomas Shulthess

Thierry Deutsch

Peter Weinberger

Mike Payne

Chris Ford

Joe Jerome

Jing Guo

Karl Hess

Joe Cychosz

Marek Turowski

Karl Rupp

Hongqi Xu

Stephen Goodnick

Jonathan Bird

Andreas Schenk

Anant Anantram

Aldo Di Carlo

Michael Rieth

Mark Ratner

Amit Chaudhry

Doris Schmitt-Landsiedel

Javier Mateos

Tomas Gonzales

Juergen Wilde

Alessandro Toscano

Pierpaolo Palestri

David Esseni

Paulette Clancy

Dragica Vasileska

Dave Ferry

Jean Michel Serrier

Frederick Green

Luciano Colombo

Mihail Nedjalkov

Siegfried Selberherr

Peter Vogl

Asen Asenov

Eleonora Alfinito

Alessandro Toscano

Abel Garcia Barrientos

Robert Baptist

Anna Perri

Juergen Sprekels

Ivan Dimov

Fausto Rossi

Lino Reggiani

Christoph Jungemann

Andreas Greiner

Enrique Velasquez

Rosana Rodriguez

Klas Lilja

Luciano Colombo

Henning Schomerus

Francis Calmon

Analysis of the perspectives of multiscale simulation of ... · Materials and device modeling It is...

Documents

Transcript of Analysis of the perspectives of multiscale simulation of ... · Materials and device modeling It is...